RUN BY JEVPlay the daily ↗

NEW YORK + LONDON / PLAYABLE NOW

Jev runs
New York.
You bring
the chaos.

Close a station. Send in a crowd.
See how Jev, the AI by TypeSafe, responds.

TODAY’S DAILY · RANKED

Same start for everyone. Your best score goes on the leaderboard.

New York daily ↗London daily ↗

Or free play: practice anytime, not ranked.New York ↗London ↗

How does Jev decide? ↓

Two real networks. No sign-up.
Trains and passengers are simulated.

NEW YORK / LIVE PREVIEWEXPLORE ↗
CONNECTING TO THE SIMULATIONWhat’s Jev’s next move?
NETWORK HEALTH —
MTA NETWORK GEOGRAPHY 40.7128° N / 74.0060° W
421 station complexes28 subway services1 AI at the controlsStep inside the network ↗

TODAY’S LEADERS

New York—

London—

See the leaderboard ↗Player stats ↗

THE CITY KEEPS MOVING. UNTIL IT DOESN’T.

One small move.
A city-wide ripple.

Every disruption spreads through the network. Jev has seconds to decide what to do with the mess.

01

Watch it decide.

Enter a shared city that keeps running. Follow Jev’s decisions on the map, one move at a time.

02

Make some trouble.

Start from this moment in your own copy of the city. You have 90 simulated seconds to put Jev to the test.

03

Send the challenge.

Share the recorded run. Your friend gets the same starting conditions—and a chance to do more damage.

INSIDE THE DECISION / EVERY 3 SIMULATED SECONDS

Read the city.
Choose a move.
See what changes.

Jev gets a short report about the simulation, a menu of possible moves and a 15-second forecast for each. Its job: reduce crowding and keep the network healthy.

  1. 01 / LOOK

    Where is the pressure?

    The game highlights crowded stations, closures, stuck trains and the reserves still available.

    ILLUSTRATIVE STATION REPORT
    150 waiting / 100 capacity

    150% full · more people than room

    Station open2 reserve trains
  2. 02 / CHOOSE

    What could help?

    Jev weighs immediate relief against help that lasts, using the available moves and their limits.

    ONE POSSIBLE CHOICESend a reserve train

    More service can help a busy, open station over time. It uses one reserve and cannot fix blocked tracks.

  3. 03 / CHECK

    Did the city recover?

    The simulation applies the move. Passengers and trains keep moving, and the next report shows what actually changed.

    WHAT GETS MEASURED
    ≋

    Station crowdingAre fewer people waiting?

    ↗

    Network healthIs the whole city holding up?

    ↻

    A fresh reportJev chooses again in 3 game seconds.

Look again. Choose again. A move is a decision, not a guaranteed rescue. We record the result so you can judge it.

See actual results ↗

Four kinds of move. Real trade-offs.

Only moves allowed by the current simulation are offered.

Send a trainMore service

Add a 90-space reserve train to a line. It carries passengers repeatedly, but costs one reserve and cannot board at a closed station.

Move passengersImmediate relief

Move up to 40 waiting passengers to a connected, less crowded station. This spreads the crowd; it does not create more trains.

Add spaceMore room per train

Add five spaces to each eligible moving train on a line, up to 150 per train. More seats help when trains fill up; they do not make trains faster.

WaitHold for now

Make no change this moment. Sensible when the city is coping; it relieves nothing, and reserves left unused at the end are worth nothing.

Real map. Simulated decisions. Jev reads game conditions, not the map image. Official service alerts and arrivals are shown separately as real-world context. Jev does not control real trains, repair signals or end your disruptions.

Create a problem. Watch Jev respond ↗

SMALL DECISIONS. MEASURABLE RESULTS.

More than
a pretty map.

Yes. Jev ran New York’s subway better than a purpose-built rulebook. +1.3 health points on average, and ahead in 48 of 48 matched tests. Each decision takes about 0.3 s and costs $0.0001.

Inspect the evidence ↗
327msMedian model response
$0.11Estimated cost / 1,000 decisions
48Matched test conditions

Recorded benchmark
Last updated:
972 recorded responses · metro-choice-4 · simulator v6 (nyc-jev-6). Inference only; failed-request usage and hosting excluded.

JEV MINUS EACH ALTERNATIVE New York · average health · 95% interval · right of the line = Jev better

Fixed rules
+1.26
Forecast-only script
+0.00
No intervention
+4.80
The honest result

Jev was consistently better than fixed rules: +1.26 points on average (95% interval +1.12 to +1.40). No reliable difference between Jev and the forecast-only script: +0.00 points on average (95% interval −0.02 to +0.02). In London, Jev was slightly but consistently better than fixed rules: +0.51 points on average (95% interval +0.35 to +0.66).

THE ROADMAP / NEXT STOPS

Two cities live.
More on the way.

One model. Different networks.
New York and London are playable. Paris is next.

01 / NOW BOARDINGPLAYABLE
Line illustration of the New York skyline

New YorkUS / SUBWAY

The first experiment. Real MTA geography, live transit context, and a simulated city you can challenge.

Play New York ↗
02 / NOW BOARDINGPLAYABLE
Line illustration of London and its clock tower

LondonUK / UNDERGROUND

272 stations. 11 Tube lines. Official TfL data and a new network to put Jev to the test.

Play London ↗
03 / NEXT STOPSCOMING SOON
Line illustration of Paris and the Eiffel Tower

ParisFR / MÉTRO

A new network. New bottlenecks. The Paris Métro is a planned next chapter for the experiment.

In the roadmap · not playable yet

Also ahead: richer transit modeling, a longer record of decisions, and better ways to compare outcomes. Plans may evolve; no launch dates promised.

Follow the build ↗

BUILT IN THE JUNGLE. OPEN TO CURIOSITY.

I wanted to see what a model built for everyday decisions could do when you gave it something complicated to run.

Thomas Kanze · independent builder ↗
THE MODEL
Jev by TypeSafe ↗

THE CITY IS WAITING.

Your move.

Play today’s daily ↗

Same start for everyone, one board per city. Or practice in free play ↗.